Scientists in Saratov have developed an innovative system that enables the identification of unmanned aerial vehicle (UAV) types based on the sounds they produce. According to the press service of Yuri Gagarin Saratov State Technical University (SSTU), where the system was created, it demonstrates high efficiency even in urban noise environments and under conditions of limited visibility.
The university emphasized that the developed system is capable of bypassing factors that make drone recognition difficult, successfully operating in a noisy urban environment and with poor visibility, and adapting to new drone models.
In the process of creating the software, the developers conducted a large-scale study aimed at identifying the weaknesses of existing drone detection systems and analyzing commercial and research solutions used in this area. As a result, a combined approach was proposed that combines methods of frequency-time sound analysis with deep machine learning algorithms based on artificial intelligence.
According to the project manager, Sergey Kuznetsov, expanding the capabilities of UAVs entails an increase in the risks associated with their unauthorized use. He noted that traditional detection methods, such as visual or radar, have drawbacks: they are ineffective in poor visibility, require high costs, and do not always provide stealth. Therefore, the development of alternative approaches, including the analysis of acoustic signals, is an urgent task.
The researcher added that the developed solution can be easily integrated into the existing infrastructure and works on standard equipment. Its use will reduce the risks of unauthorized use of UAVs due to prompt detection. Testing confirmed the effectiveness of the system in urban conditions, where it successfully identifies drone acoustic signals against the background of extraneous noise.
The university reported that data augmentation methods were used to train the system, including adding noise and changing speed, as well as transfer learning. This ensured high recognition accuracy even in the presence of background interference.
The scientific supervisor of the project was Mikhail Korolev, PhD in Engineering, Associate Professor of the Department of Applied Information Technologies at SSTU.